Most HIV infections originate from individuals who are undiagnosed and unaware of their infection. Estimation of this quantity from surveillance data is hard because there is incomplete knowledge about i) the time between infection and diagnosis (TI) for the general population and ii) the time between immigration and diagnosis for foreign-born persons. Development: We developed a new statistical method for estimating the number of undiagnosed people living with HIV (PLHIV) and the incidence of

more »

... IV-1 based on dynamic modeling of heterogenous HIV-1 surveillance data. We formulated a Bayesian non-linear mixed effects model using multiple biomarkers to estimate TI accounting for biomarker correlation and individual heterogeneities. We explicitly model the probability that an HIV-1 infected foreign-born person was infected either before or after immigration to distinguish between endogenous and exogeneous incidence. The incidence estimator allows for direct calculation of the number of undiagnosed persons. Application: The model was applied to surveillance data in Sweden. The dynamic biomarker model was trained on longitudinal data from 31 treatment-naive patients with well-defined TI, using CD4 counts, BED serology, polymorphisms in HIV-1 pol sequences, and testing history. The multiple-biomarker model was more accurate than single biomarkers (mean absolute error 1.01 vs >= 1.95). We estimate that 813 (95% CI 780-862) PLHIV were undiagnosed in 2015, representing a proportion of 10.8% (95% CI 10.4-11.3%) of all PLHIV. Conclusions: The proposed methodology will enhance the utility of standard surveillance data streams and will be useful to monitor progress towards and compliance with the 90-90-90 UNAIDS target. Key words: HIV-1, BED assay, pol sequences, incidence estimation, undiagnosed HIV-1 infections.